Research Article
A Practical Deep Learning Model in Differentiating Pneumonia-Type Lung Carcinoma from Pneumonia on CT Images: ResNet Added with Attention Mechanism
Table 2
Evaluation indexes of overall effectiveness of the radiologist, deep learning, and radiologist joint model in diagnosing pneumonia-like lesions.
| | | Junior radiologist | Senior radiologist | Model | Junior radiologist + model | Senior radiologist + model |
| | No. of correct diagnosis | 25/36 | 27/38 | 32/42 | 32/44 | 33/45 | | LLF | 61% | 65% | 74% | 76% | 78% | | NLF | 32.4%
(95% CI: 0.19–0.50) | 27.01%
(95% CI: 0.14–0.44) | 13.51%
(95% CI: 0.05–0.30) | 13.51%
(95% CI: 0.05–0.30) | 10.81%
(95% CI: 0.04–0.26) | | Sensitivity | 48.0%
(95% CI: 0.34–0.62) | 51.92%
(95% CI: 0.38–0.66) | 60.37%
(95% CI: 0.46–0.73) | 62.75%
(95% CI: 0.48–0.76) | 64.71%
(95% CI: 0.50–0.77) | | Specificity | 75.0%
(95% CI: 0.60–0.86) | 79.17%
(95% CI: 0.65–0.89) | 89.36%
(95% CI: 0.76–0.96) | 89.80%
(95% CI: 0.77–0.96) | 91.84%
(95% CI: 0.80–0.97) | | PPV | 67.5%
(95% CI: 0.50–0.81) | 72.97%
(95% CI: 0.56–0.86) | 86.49%
(95% CI: 0.70–0.95) | 86.49%
(95% CI: 0.70–0.95) | 89.19%
(95% CI: 0.74–0.96) | | NPV | 57.1%
(95% CI: 0.44–0.69) | 60.31%
(95% CI: 0.47–0.72) | 66.67%
(95% CI: 0.54–0.78) | 69.84%
(95% CI: 0.57–0.80) | 71.43%
(95% CI: 0.58–0.82) |
|
|
